Neural Assemblies Learn Causal Directionality via DIRECT Mechanism

Researchers have revealed that neural assemblies—clusters of neurons that activate simultaneously and strengthen through co-activation—are capable of learning the causal influence direction between variables, a phenomenon not previously demonstrated. The study presents DIRECT (DIRectional Edge Coupling/Training), a method that co-activates source and target assemblies using an adaptive gain schedule to internalize directed relationships. In contrast to methods based on backpropagation, DIRECT operates exclusively on local plasticity, allowing for auditable causal claims at the mechanism level. The findings are confirmed through a dual-readout validation approach that examines synaptic-strength asymmetry. This research positions neural assemblies as a broadly applicable framework for causal learning, enhancing their established roles in classification, parsing, and planning.